Strategy-proof allocation of indivisible goods

Soc Choice Welfare (1999) 16: 557±567 Strategy-proof allocation of indivisible goods Lars-Gunnar Svensson Department of Economics, Lund University, P.O. Box 7082, SE-220 07 of Lund, Sweden (e-mail: lars-gunnar.svensson@nek.lu.se) Received: 15 September 1997/Accepted: 12 May 1998 Abstract. Strategy-proof allocation of a nite number of indivisible goods among a nite number of individuals is considered. The main result is that in a pure distributional case, a mechanism is strategy-proof, nonbossy and neutral if and only if it is serially dictatorial. If the indivisible goods are initially owned by the individuals, a mechanism is strategy-proof, individually rational and Pareto consistent if and only if it is the core mechanism. 1 Introduction This study will determine the set of strategy-proof allocation mechanisms in a model with a nite number of indivisible goods to be allocated to a nite number of individuals. In the classical study of this type of model, Shapley and Scarf (1974), had houses in mind and the existence of a competitive equilibrium was established. Roth and Postlewaite (1977) showed that the core in this model contains only one allocation if preferences are strict rankings, while Roth (1982) proved that the core could be implemented by a strategy-proof allocation mechanism. Finally Ma (1994) showed that there is only one strategy-proof, individually rational and Pareto optimal1 allocation mechanism in this model and that the outcome of the mechanism is the unique core allocation. This mechanism also has the property that no subset of individuals can improve the outcome of the mechanism for the group by coordinating and misrepresent their preferences (see Bird 1984). * I would like to thank Bettina Klaus and two anonymous referees for their very helpful comments on this paper. Financial support from The Bank of Sweden Tercentenary Foundation is gratefully acknowledged. 1 Pareto consistent in our terminology.

558 L.-G. Svensson Here the corresponding problem is examined but without an initial property right structure. For example, it is the government that has to distribute the indivisible goods among the individuals. We may still interpret the indivisible goods as houses or jobs, or as in Hylland and Zeckhauser (1979), as various positions with limited capacities. The basic assumption is that the individuals demand only one of the indivisible goods. The main result of this paper is the complete characterization of the set of strategy-proof allocation mechanisms under the additional assumptions of neutrality and nonbossiness. In that case the allocation mechanisms must be serially dictatorial, i.e. every mechanism determines a queue order, an order in which the individuals have to choose an indivisible good from the set of ``remaining'' goods when individuals with higher rankings have made their choice2. The existence of a serial dictatorship is also found in Satterthwaite and Sonnenschein (1981) in a model with divisible goods. Our strategy-proof result is very similar to the Gibbard-Satterthwaite theorem (Gibbard 1973; Satterthwaite 1975). If the indivisible goods are interpreted as public goods instead of private goods as in the present paper, the allocation mechanism is a ``voting procedure'' and serial dictatorship is replaced by dictatorship. In the case of public goods, nonbossiness is automatically satis ed, while neutrality in our model is there replaced by the voting procedure being ``onto''3. We will also reconsider the model with property rights and give a simple and alternative proof of the result in Ma (1994); a mechanism is strategyproof, individually rational and Pareto consistent if and only if it is the core mechanism. It is interesting that individual rationality and preferences together determines a queue order by which the individuals have to choose an indivisible good. The di erence from the nonproperty rights case is that the queue order becomes endogenous. Strategy-proof exchange in a neoclassical model with divisible goods is analyzed in BarberaÁ and Jackson (1995). In that case strategy-proofness requires trade to occur according to a number of pre-speci ed proportions, but there is no obvious relationship with the indivisible goods case. Strategy-proof and e½cient allotment rules are also characterized in BarberaÁ, Jackson and Neme (1997), where the agents have to share a divisible good (or task). Their characterization leads to a class of mechanisms called ``sequential allotment rules'', which to some extent resemble a serial dictatorship. More closely related works to the present study are Papai (1996, 1998a,b). In these studies, three di erent models with indivisible goods are considered and a complete characterization of the set of strategy-proof, Pareto-optimal 2 In Svensson (1994) it is demonstrated that queue allocations may be implemented by a strategy-proof allocation mechanism also in the case when indi erences in the preferences are permitted. 3 If a voting procedure is assumed to be neutral (or Pareto consistent) instead of onto, a very short and simple proof of the Gibbard-Satterthwaite theorem can be obtained, see Svensson (1997).

Strategy-proof allocation 559 and nonbossy social choice functions is given. In the present study neutrality is an essential assumption and serial dictatorship is a consequence, while Papai (1998b) assumes Pareto optimality and receives a larger set of mechanisms called hierarchical exchange functions. Hence, within this framework our neutrality condition is a more restrictive assumption than Pareto optimality. There are also two other papers, SoÈ nmez (1995) and Abdulkadiroglu and SoÈ nmez (1997), analyzing models with indivisible goods like the present study. In the rst paper the relationship between strategy-proofness and uniqueness of the core is examined (in a more general matching model than ours). As a corollary to his general result, SoÈ nmez obtained Ma's (1994), and hence also our result ( Theorem 2). The second paper proves the equivalence between two di erent ways of allocating the indivisible goods; serial dictatorship with a randomly determined order of choice or using the core correspondence after randomly distributing initial endowments. There are no directly overlapping results in that paper and in the present one, but the problems studied are closely related to each other4. 2 The model and de nitions Let N ˆf1; 2;...ng be a nite set of natural numbers, which also denotes a set of individuals, and let A ˆfa 1 ; a 2 ;...a m g; m V n, be a nite set of indivisible goods. We may interpret the elements in A as houses, jobs, occupations, positions etc. Preferences over A are rankings of the various elements (i.e. complete, transitive and asymmetric binary relations). The set of all possible rankings of A represented by utility functions is denoted U, so for u A U and a; b A A with a 0 b; u a > u b or u b > u a, but not both. Preference pro les are elements in U ˆ U n. A preference pro le u ˆ u 1 ; u 2 ;...u n can also be denoted u i ; u i for i A N, or u S ; u S for S H N, where u S contains utility functions u i with i A S and u S contains the rest of the utility functions. An allocation is an injective mapping j from N to A. For a given preference pro le u A U, an allocation j is Pareto e½cient if there is no other allocation c such that for all i A N, u i c i V u i j i, with strict inequality for some i A N. Moreover, for a given preference pro le u A U, j is a queue allocation if there is a permutation p of N (a queue order) such that j i j is the best element in the set A fj i 1 ; j i 2 ;...j i j 1 g, where p i j ˆj for all j A N, according to individual preferences u ij. Here p i denotes the order in which the individuals have to choose an indivisible good from the set A; the individual i with p i ˆ1 has to choose rst from A, then the individual with p i ˆ2 from the remaining part of A, and so forth. An allocation mechanism, or for short a mechanism5, is a mapping f from 4 The results in the present study and to some extent, the overlapping results in Papai (1998b) and SoÈnmez (1995), are independently obtained. 5 An alternative name of ``allocation mechanism'' in this context is social choice function.

560 L.-G. Svensson U to the set of allocations, i.e. every preference pro le is mapped on an allocation of the indivisible goods. A mechanism f is Pareto consistent if f u is Pareto e½cient for all u A U. The mechanism is manipulable precisely when there is an individual i A N, preferences v i A U, and a preference pro le u A U, such that u i f i v i ; u i > u i f i u. If a mechanism is not manipulable, it is strategy-proof. If, for a given permutation p of N, the outcome allocation f u of a mechanism f is always the corresponding queue allocation, the mechanism is called serially dictatorial. One easily shows that such a mechanism is strategy-proof, see e.g. Svensson (1994). However, the main object of this paper is to show that under some additional assumptions, nonbossiness and neutrality, every strategy-proof mechanism is serially dictatorial. A mechanism f is nonbossy6 if for all preferences v i A U and preference pro les u A U, f v i ; u i ˆ f u when f i v i ; u i ˆ f i u. Hence, nonbossiness entails that an individual cannot change the outcome of the mechanism without changing the outcome for himself at the same time. Now let p be a permutation of A ± a change of names of the indivisible goods7. If j is an allocation, an allocation pj is de ned as (pj i ˆp j i ; i A N. Ifu A U is a preference pro le, a preference pro le pu is de ned as (pu i a ˆu i p 1 a for i A N and a A A. A mechanism f is neutral if, for all preference pro les u A U and permutations p of A, f pu ˆp f u. This means that the ``real'' outcome of a neutral mechanism is independent of the names of the indivisible goods. Before a change of names an individual i receives f i u if the reported utility pro le is u. After a change of names by p the same individuals will report the pro le pu and the outcome for i is f i pu. The original name of this good is p 1 f i pu and therefore p 1 f i pu ˆ f i u for a neutral mechanism. 3 The set of strategy-proof mechanisms Before characterizing the set of strategy-proof mechanisms we will give an example of a mechanism that is nondictatorial, strategy-proof, nonbossy and Pareto consistent. The mechanism is, however, not neutral. In the rest of the section neutrality is assumed and our main result becomes that strategyproofness implies (serial) dictatorship. That result is proved with the help of two lemmas. Example. Let N ˆf1; 2; 3g and A ˆfa; b; cg. Let f be a mechanism de ned so that if a is the best element in A according to preferences u 2, then f 1 u is the best element in fb; cg according to preferences u 1, f 2 u ˆa and f 3 u is the remaining element. In all other cases f 1 u is the best element in A according to preferences u 1, f 2 u is the best element in A ff 1 u g according to 6 The concept of nonbossiness is due to Satterthwaite and Sonnenschein (1981). 7 Strictly speaking, p is a permutation of the set f1; 2;...mg of indices of the elements in A. However, for a j A A we will write p a j instead of a p j.

Strategy-proof allocation 561 preferences u 2 and f 3 u is the remaining element. Hence the mechanism f is serially dictatorial for all utility pro les except for those where individual 2 has a as the best element. (We can think of individual 2 as the owner of a.) This mechanism is obviously not neutral ± the element a has a special position ± but it is strategy-proof (individual 2 cannot increase his utility by not telling the truth!). It is also nondictatorial since no individual always gets his best element, and it is nonbossy, since an individual cannot change the outcome in general without changing the outcome of the mechanism for the individual himself. Finally, the outcome of the mechanism is obviously always Pareto e½cient. So if we can accept nonneutral mechanisms we can also nd nondictatorial mechanisms. But as is shown in the sequel, a neutrality condition reduces the set of possible mechanisms drastically. The main result in this section is Theorem 1 below. Two lemmas, which also have some interest of their own, will simplify the proof. The rst lemma shows that a strategy-proof and nonbossy mechanism is constant in certain subsets of U. A similar result can be found in BarberaÁ and Jackson (1995) or Papai (1996) but in di erent models. In the second lemma it is proved that in the subset of U where all individuals' utility functions coincide the outcome of a strategy-proof, nonbossy and neutral mechanism is always Pareto e½cient. Lemma 1. Let f be a strategy-proof and nonbossy mechanism, u; v A U two preference pro les such that for x A A and i A N, v i x U v i f i u if u i x U u i f i u. Then f v ˆ f u. Proof. We rst prove that f v i ; u i ˆ f u. From strategy-proofness it follows that u i f i v i ; u i U u i f i u and hence from the assumption of the lemma, v i f i v i ; u i U v i f i u. But strategy-proofness also implies that v i f i v i ; u i V v i f i u and then, because preferences are strict, f i v i ; u i ˆ f i u. Finally, nonbossiness implies that f v i ; u i ˆ f u. For the rest of the proof let u p ˆ v 1 ; ::: v p 1 ; u p ; ::: u n A U, i.e. u p 1 ˆ v p ; u p p, be a sequence of preference pro les. From the rst part of the proof it then follows that f u p ˆf v p ; u p p ˆ f up 1. But then f v ˆ f u because f v ˆ f u n 1 and f u ˆ f u 1. 9 Lemma 2. Let f be a strategy-proof, nonbossy and neutral mechanism, and u A U a preference pro le where all individual preferences coincide, u i ˆ u j for all i; j A N. The outcome of the mechanism is then Pareto e½cient. Proof. Assume that f u is not Pareto e½cient. There is then a A A ff i u ; i A Ng such that for some k A N, u i f k u < u i a. Also let k be chosen so that u i f k u ˆ maxfu i f j u ; u i f j u < u i a g j and let b ˆ f k u. Now de ne individual preferences v i A U, equal for all i A N, such that v i x ˆu i x for all x A A fa; bg, while v i a ˆu i b and v i b ˆu i a.

562 L.-G. Svensson Hence v i a < u i a and v i b > u i b : We can then show that the preference pro les u and v satisfy the assumptions in Lemma 1. Let for i A N, x A A be such that u i x U u i f i u. First consider x 0 b. The inequality v i x U u i x follows from the de nition of v i. Hence, v i x U u i x U u i f i u U v i f i u, where the last inequality follows because a 0 f i u. Next assume that x ˆ b. Ifi ˆ k we have directly that v i b ˆv i f i u. If i 0 k, the inequality u i b < u i f i u implies that u i a < u i f i u because of the maximal choice of f k u. But then v i f i u ˆ u i f i u. Thus we have v i b ˆu i a < u i f i u ˆ v i f i u. From Lemma 1 it now follows that f u ˆ f v. Finally let p be a permutation of A such that p x ˆx if x A A fa; bg while p a ˆb and p b ˆa. Thus u ˆ pv, and hence by neutrality, p f v ˆ f pv ˆf u. This is a contradiction to f u ˆ f v, so the statement in the lemma is true. 9 To prove the theorem below we rst consider the outcome of f when all individuals have the same preferences. In that way the con ict in the allocation situation is large and it is possible to nd directly the order in which the individuals have to choose elements from A. Next we apply Lemma 1 to prove that the order of choices remains the same in the entire set U of preference pro les. Theorem 1. A strategy-proof, nonbossy and neutral mechanism f is serially dictatorial. Proof. Let u A U be a preference pro le where all individual preferences coincide and u i a j > u i a k when k > j for all i A N. From Lemma 2 follows then that f f i u ; i A Ng ˆfa 1 ;...a n g. Now there is a permutation s of N such that f s j u ; j A N, is the best element in A ff s 1 u ;... f s j 1 u g according to the common preferences. With no loss of generality we may assume that s i ˆi and f i u ˆa i for all i A N. Hence, for this particular preference pro- le, individual 1 is the dictator and chooses from the entire set A, individual 2 chooses from the remaining alternatives, and so forth, and the mechanism is serially dictatorial. From neutrality it easily follows that this order of choice is the same for all preference pro les where all individual preferences coincide. To see this let v A U and v i ˆ v j for all i; j A N. There is then a permutation p of A such that u i a ˆv i p a for all a A A and i A N. But then pu ˆ v and hence by neutrality, p f u ˆ f pu ˆ f v. This entails that the utility distributions are the same for all preference pro les where all individual preferences coincide because v i f i v ˆ v i f i pu ˆ v i p f i u ˆ u i f i u : Thus individual 1 is the dictator, individual 2 is the second one to choose, and so forth. We will nally show that this order of choice remains for all preference pro les.

Strategy-proof allocation 563 Let v A U be an arbitrary preference pro le and de ne recursively a subset fa ij g n jˆ1 of A according to: a ij is the best element in A fa i1 ; a i2 ;...a ij 1 g according to v j ; i.e. the elements a ij are the outcome of a serial dictatorship where individual 1 makes the rst choice, individual 2 is the second one to choose, and so forth. Let u A U be a preference pro le where all individual preferences coincide, u i ˆ u j for all i; j A N, and satisfy: u i a ij > u i a ik if j < k U n; u i a in > u i a for all a A A fa i1 ; a i2 ;...a in g: From the rst part of the proof it follows that f j u ˆa ij for all j A N, i.e. individual 1 is the dictator etc. It remains to show (by Lemma 1) that f u ˆ f v. Let a A A and u j a U u j f j u. Then a A A fa i1 ; a i2 ;...a ij 1 g. From the de nition of fa ij g n jˆ1 follows then that v j a U v j a ij ˆv j f j u and hence by Lemma 1, f u ˆ f v. 9 Theorem 1 shows that if there is a strategy-proof, nonbossy and neutral mechanism it must be serially dictatorial. But one easily shows that the serially dictatorial mechanism also satis es those conditions, so the set of such mechanisms is nonempty. In the example in the beginning of this section we demonstrated that neutrality was important for the result in Theorem 1; strategy-proof, nonbossy and Pareto consistent mechanisms that are not dictatorial do exist. A related problem is whether the nonbossy requirement is necessary for the result. However, without nonbossiness a strategy-proof and neutral mechanism must not necessarily be serially dictatorial. For instance, the ``pure'' dictatorial dictatorship where f i u is the best element in A ff 1 u ; f 2 u ;... f i 1 u g according to individual 1's preferences u 1 is strategy-proof and neutral but not Pareto consistent. Suppose then that nonbossiness in Theorem 1 is replaced by Pareto consistency. The following example shows that serial dictatorship is not the implication. Let an ˆ aa ˆ 4 and consider a mechanism that is ``almost'' serially dictatorial; the order of choice is individual 1, 2, 3 and 4 in all cases except when individual 1 and 2 report the same preferences. In that case individual 4 makes his choice before individual 3. It is easy to show that this mechanism is strategy-proof, neutral and Pareto consistent, but it is also bossy. I do not have, however, a complete characterization of mechanisms that are strategy-proof, neutral and Pareto consistent8. In next section we will assume that each alternative in A is owned by precisely one individual and hence we cannot assume neutrality. Nonbossiness will also be replaced by Pareto consistency. 8 The set of strategy-proof, nonbossy and Pareto consistent mechanisms are characterized in Papai (1998b).

564 L.-G. Svensson 4 Strategy-proofness with property rights Now suppose that m ˆ n and that each individual owns precisely one indivisible good. With no loss of generality we assume that individual i A N owns a i A A. Let S H N be a nonempty set of individuals (a coalition), u i A U; i A S, utility functions, and j an allocation. Then the coalition S blocks the allocation j if there is another allocation c such that c i ˆa p i for all i A S for some permutation p of S, and such that u i c i V u i j i for all i A S with strict inequality for some i A S. An allocation j is individually rational if no coalition with one member blocks j, Pareto e½cient if the entire coalition N does not block j, and in the core9 if no coalition blocks j. Moreover, a mechanism f is individually rational if the allocation f u is individually rational for all preference pro les u A U, Pareto consistent if f u is Pareto e½cient for all preference pro les u A U, and a core mechanism if f u is in the core for all preference pro les u A U. Let j be an allocation and S H N; S 0ja nonempty subset of individuals. The set j S ˆfj i ; i A Sg is a cycle if S ˆfp k j ; k ˆ 1;...ng for some j A S, where p is the (unique) permutation of N given by j p i ˆ a i for all i A N. Hence, the element a j is allotted to individual p j A S, the element a p j is allotted to individual p 2 j ˆp p j A S, the element a p 2 j is allotted to individual p p 2 j ˆ p 3 j A S etc. Note that j i 0 a i for all i A S if as > 1. Lemma 3. For each u A U there is precisely one core allocation j. There is also a corresponding partition10 fn j g r jˆ1, r U n, of the set N such that A j ˆ j N j are cycles for all j; 1 U j U r; and j 1 if i A N j then u i j i V u i a for all a A A 6 A k : kˆ1 Proof. The uniqueness of the core allocation j follows from Roth and Postlewaite (1977). The sets fn j g r jˆ1 are de ned sequentially. First de ne N 1. Let r : N! N be a mapping such that for i A N, a r i is the best element in A according to u i. Because N is nite there is a set fi 1 ; i 2 ;...i k g H N such that i j 0 i l if j 0 l; i j 1 ˆ r i j and i 1 ˆ r i k. Let N 1 ˆfi 1 ; i 2 ;...i k g. Hence N 1 ˆfr q i 1 ; q ˆ 1; ::: ng. Since j is the unique core allocation j N 1 ˆfa r i ; i A N 1 g. Obviously j N 1 is a cycle with r 1 i ˆp i for i A N 1 and j p i ˆ a i. Denote by A 1 the set j N 1. From the construction of N 1 also follows that u i j i V u i a for all a A A if i A N 1. To de ne N 2 and A 2 we repeat the procedure in the sets N N 1 and A A 1, and so forth for the rest of the sets. 9 9 The core de ned by weak domination like here is also called the strict core. 10 One can also show that the partition is unique as far as the numeration of the sets N j, but that is not needed for the proof of the next theorem.

Strategy-proof allocation 565 The existence of a strategy-proof, individually rational and Pareto consistent mechanism is proved in Roth (1982). Here we give an alternative proof of Ma's (1994) result that the core mechanism is the only one. A more simple proof is obtained if in addition nonbossiness is assumed. That proof is given in an appendix. From Lemma 3 we may also note that a core allocation is a queue allocation where the queue order is endogenously determined by the preference pro le. This is in contrast to the pure distributional case in the preceding section where strategy-proof mechanisms were characterized by an exogenously given queue order. Theorem 2. A strategy-proof, individually rational and Pareto consistent mechanism f is the core mechanism. Proof. Let u A U and let j be the unique corresponding core allocation. Also let fn j g r jˆ1 and fa jg r jˆ1 ; r U n, be partitions of the sets N and A in accordance with Lemma 3. To simplify notation, we may assume with no loss of generality that i > k if i A N j, k A N l and j > l. We rst prove that for all i A N 1, f i u N1 ; u N 0 1 ˆj i for all preference pro les u N 0 1 for the set N N 1 of individuals. If an 1 ˆ 1 then by individual rationality, f 1 u N1 ; u N 0 1 ˆj 1 ˆa 1. On the other hand, if an 1 > 1thenfor i A N 1, j i 0 a i because j N 1 is a cycle. In this case, let N 1 ˆf1; 2;...kg. For i A N 1 de ne utility functions v i A U according to v i a ˆu i a for all a A A fa i g; while for a i ; u i j i > v i a i > u i a for all a A A fa i ; j i g: In other words, with preferences v i, initial resources are always second best while the ranking of other elements are the same in v i and u i. Also denote by v p N 1 preference pro les for the set N 1 such that there are p individuals having the original utility functions u i while the rest of the individuals in N 1 have utility functions v i, e.g. v p N 1 ˆ v 1 ; u 2 ;...u p 1 ; v p 2 ;...v k. Then by Pareto consistency and individual rationality, f i vn 0 1 ; u N 0 1 ˆj i for all i A N 1. Suppose now that for i A N 1 ; f i v p N 1 ; u N 0 1 ˆj i for all v p N 1 for a xed number p, 0U p < k. Then f i v p 1 N 1 ; u N 0 1 ˆj i for all i A N 1 having utility functions u i by strategy-proofness and hence by individual rationality, f i v p 1 N 1 ; u N 0 1 ˆj i also for the rest of the individuals in N 1. Then by induction, f i u N1 ; u N 0 1 ˆj i for all i A N 1 because u N1 ˆ vn k 1. To complete the proof we consider the sets N N 1 and A A 1 and repeat the procedure above to prove that for i A N 2, f i u N1 ; u N2 ; u N 0 1 W N 2 ˆj i for all u 0, and so forth for the rest of the sets N j and A j. 9 5 Appendix Theorem 3. A strategy-proof, individually rational, nonbossy and Pareto consistent mechanism f is the core mechanism.

566 L.-G. Svensson Proof. Let u A U and let j be the unique corresponding core allocation. Also let fn j g j and fa j g j be partitions of the sets N and A in accordance with Lemma 3 in section 4. To simplify notation, we may assume with no loss of generality that i > k if i A N j, k A N l and j > l. Now de ne recursively a preference pro le v A U by: For i A N j de ne utility functions v i A U according to v i ˆ u i if j i ˆa i, while if j i 0 a i, v i a ˆu i a for all a A A fa i g while for a i ; j 1 u i j i > v i a i > u i a for all a A A 6 A k! fa i ; j i g: kˆ1 But then by Pareto consistency and individual rationality, f v ˆj according to the construction of v. To determine f u rst consider f v p, where v p ˆ u 1 ; u 2 ;...u p 1 ; v p ;...v n A U. Obviously v 1 ˆ v and v n 1 ˆ u. Suppose that f v p ˆj. Then f p v p 1 ˆj p by strategy-proofness and then by nonbossiness, f v p 1 ˆj. Hence f u ˆj by induction. 9 References [1] Abdulkadiroglu A, SoÈnmez T (1997) Random Serial Dictatorship and the Core from Random Endowments in House Allocation Problems. Econometrica (forthcoming) [2] BarberaÁ S, Jackson MO (1995) Strategy-Proof Exchange. Econometrica 63: 51±87 [3] BarberaÁ S, Jackson MO, Neme A (1997) Strategy-Proof Allotment Rules. Games Econ Beh 18: 1±21 [4] Bird CG (1984) Group incentive compatibility in a market with indivisible goods. Econ Lett 14: 309±313 [5] Gibbard A (1973) Manipulation of Voting Schemes: A General Result. Econometrica 41: 587±601 [6] Hylland A, Zeckhauser R (1979) The e½cient allocation of individuals to positions. J Polit Econ 87: 239±314 [7] Ma J (1994) Strategy-Proofness and the Strict Core in a Market with Indivisibilities. Int J Game Theory 23: 75±83 [8] Papai S (1996) Strategyproof and Non-Bossy Assignments. Social Science Working Paper No. 938, California Institute of Technology [9] Papai S (1998a) Strategyproof Single Unit Award Rules. Working Paper Series No: 1998/2, Koc University, mimeo [10] Papai S (1998b) Strategyproof Assignment by Hierarchical Exchange. Department of Economics, Koc University, mimeo [11] Roth AE (1982) Incentive compatibility in a market with indivisible goods. Econ Lett 9: 127±132 [12] Roth AE, Postlewaite A (1977) Weak versus strong domination in a market with indivisible goods. J Math Econ 4: 131±137 [13] Satterthwaite MA (1975) Strategy-proofness and Arrow's Conditions: Existence and Correspondence Theorems for Voting Procedures and Social Welfare Functions. J Econ Theory 10: 187±216 [14] Satterthwaite MA, Sonnenschein H (1981) Strategy-Proof Allocation Mechanisms at Di erentiable Points. Rev Econ Studies 48: 587±597 [15] Shapley L, Scarf H (1974) On Cores and Indivisibility. J Math Econ 1: 23±37

Strategy-proof allocation 567 [16] SoÈnmez T (1995) Strategy-Proofness and Singleton Cores in Generalized Matching Problems. Department of Economics, University of Michigan, mimeo [17] Svensson L-G (1994) Queue allocation of indivisible goods. Soc Choice Welfare 11: 323±330 [18] Svensson L-G (1997) A Simple Proof of the Gibbard-Satterthwaite Theorem. Working paper, Department of Economics, Lund University